Runway Aleph 2.0 starts from a video that already exists. Instead of asking a model to invent a whole clip again, a creator can change a product, background, season, lighting condition, object, or visual style while attempting to preserve the action and everything outside the edit.

Runway launched Aleph 2.0 and its accompanying Edit Studio on May 21, 2026. The model accepts clips up to 30 seconds at 1080p and is available in the desktop web app on all paid Runway plans. Its central product idea is image-level control over a moving sequence: define the intended result on a frame, then carry that visual choice through the relevant parts of the video.

That makes Aleph different from both a conventional timeline editor and a text-to-video generator. It is generative because the model synthesizes changed pixels across time. It is an editor because the existing footage supplies composition, performance, camera movement, timing, and most of the scene.

The preservation problem in generative editing

An apparently simple instruction such as “make the shoes red” has two requirements. The shoes must change, and everything else must not. A model can satisfy the first while quietly changing the actor’s face, background objects, framing, or motion. For commercial footage, those collateral changes can make an otherwise impressive result unusable.

Runway positions Aleph 2.0 around localized edits with stronger input-video preservation. The model is intended to modify the requested element while retaining scene action, cuts, lighting, and unrelated objects. Its product page gives examples such as changing a product color, hairstyle, clothing, background, season, or overall style and removing an unwanted item.

Preservation is still a model objective, not a mathematical lock. Fine details can drift, occluded objects can reappear, and changes can interact with reflections or shadows. A production review should compare the output with the original, not merely ask whether the requested edit appears.

Frame-guided control

Edit Studio lets the creator preview the edit as an image before generating the full video. That still frame becomes a concrete visual target. Instead of relying entirely on the model’s interpretation of a sentence, the editor can refine the look until the product, palette, or environment is close to the intended result.

The workflow is straightforward:

  1. upload or select the source video;
  2. describe the desired change;
  3. preview the result on a representative frame;
  4. refine the frame until the target is clear;
  5. generate the temporal edit;
  6. inspect the result across the full clip.

This separates two expensive questions: “What should the changed scene look like?” and “Can that look remain coherent through motion?” A failed frame preview is cheaper to reject than a failed 30-second render. The approach also creates a useful approval point for brand teams before compute is spent on the final video.

The chosen frame should expose the difficult part of the edit. If a product is visible only from the front in the preview but rotates later, the target says little about its side or back. Reviewers should sample moments with occlusion, motion blur, camera changes, and contact with hands or surfaces.

Editing across several shots

Aleph 2.0 can apply an edit across relevant shots in a video containing cuts or scene changes. A campaign clip may show the same package in a close-up, a wide shot, and a final product tableau. Editing all three at once is more useful than treating each shot as an unrelated generation.

Multi-shot support also raises the quality bar. The new package must remain recognizable across scale and lighting changes. A replacement object should not adopt one design in the close-up and another in the wide shot. Irrelevant shots should remain untouched.

Create a shot-level review checklist before generating. Record where the target appears, what must stay fixed, and which shots should not be edited. For longer assets, splitting the work into approved sections may provide better control than submitting the full 30 seconds simply because the limit allows it.

Where Aleph can save a reshoot

The strongest use cases change a bounded visual property while preserving a valuable performance or camera move. Examples include:

These edits can reduce the need to reassemble a cast, location, product, and crew. They do not eliminate every conventional post-production task. Precise legal text, frame-accurate graphics, audio mixing, accessibility captions, and delivery specifications still benefit from deterministic tools.

The source footage also matters. Heavy compression, fast motion, transparent materials, reflections, and partial occlusion make object identity difficult to maintain. Plan a small feasibility test before promising a complete campaign conversion.

Availability, cost, and operational limits

Aleph 2.0 is available through Edit Studio on paid Runway plans. Runway’s announcement does not publish one universal per-edit price; usage depends on the current plan and credit model. Teams should check the live pricing page and record actual credits per accepted output rather than preserving a launch-day estimate in permanent documentation.

The 30-second, 1080p limit is suitable for ads and social clips, not an entire long-form program. Longer work must be divided, creating additional continuity and assembly requirements. Edit Studio is a desktop web experience at launch, and the model’s availability through other Runway surfaces may differ.

Generative retries are the main hidden cost. A precise frame preview can reduce them, but it cannot guarantee temporal success. Keep the original, preview frame, prompt, output, and review notes together so a team can compare attempts rather than regenerating from memory.

A production review for localized edits

Review the full output at normal speed and frame by frame around transitions. Check:

For product work, compare against an approved reference, not only the prompt. For people, obtain permission for material changes and avoid edits that misrepresent an individual’s actions or endorsement.

Generative editing followed by deterministic finishing

Aleph is best used for the creative visual change it was designed to make. Once that change is approved, a separate finishing layer can handle operations whose expected output is mechanically defined. Medux provides several such external media tools that Codex or Claude can call through an independently configured MCP connection; this does not mean Runway and Medux are natively integrated.

For example, an authorized source may contain a visible subtitle band that must be removed before the Aleph edit. The Medux subtitle-removal tutorial for Codex MCP shows upload, asynchronous task creation, polling, and download. The result can then enter the generative edit, with the usual visual review for reconstructed pixels.

After Aleph, the approved clip may need a mastered soundtrack and fixed brand mark. The Claude MCP audio-track workflow combines a video and audio file, while the Codex MCP logo tutorial applies a supplied watermark at a chosen position. Those operations should not reinterpret the scene.

A controlled sequence is:

  1. preserve and hash the original source;
  2. perform only necessary deterministic cleanup;
  3. create and approve the Aleph frame target;
  4. generate the visual edit and inspect every shot;
  5. add final audio and branding from approved assets;
  6. verify duration, dimensions, audio, logo placement, and output filename;
  7. require a final rights and publication review.

Aleph 2.0 narrows the distance between existing footage and a new creative version. The most reliable pipeline keeps generative judgment focused on that visual transformation and leaves fixed delivery operations to tools that can be validated exactly.